Stop pushing broken code to CI: Wire Chunk sidecars into agent hooks

AI agents, while capable of writing code rapidly, often face a disconnect in the feedback loop, as errors might not be detected until after the code reaches the CI pipeline. To address this, Chunk sidecars introduce a mechanism that intercepts and corrects failures before they proceed to CI by using lightweight Linux microVMs that mirror the CI environment. These sidecars run test suites quickly, allowing agents to fix simple errors within their workflow, thus reserving the CI pipeline for integration and deployment tasks. The process involves agent hooks, which are scripts that trigger automated validation at specific checkpoints, ensuring that errors are caught and corrected promptly. Two main hooks are used: a "commit gate" that checks code before committing by running local tests, and a "finish gate" that conducts a comprehensive validation in a remote environment before the agent completes its task. This system not only enhances the efficiency of the feedback loop but also ensures that the CI pipeline is not delayed by easily preventable failures, thereby optimizing the development process and maintaining code quality.